GateTrue

What each generator documents about marking its output

What earns a row on this site

What earns a row here, and what would make this a different siteA statement earns a row when it appears on a page the vendor controls. Looking at output to determine whether a file is marked would be a different kind of site, producing findings rather than records. One vendor per page, no aggregate scores, and omissions recorded as omissions.A claim about markingOn the vendor's pageEarns a rowQuoted, dated, and linkedto where it was read.Observed in outputOut of scopeThat would be testing, andthis site does not test.NowhereRecorded as anomissionThe empty cell is therecord.One vendor per page, and no aggregate scores
Fig. 1 The boundary is between what a vendor says and what a file contains. Crossing it would mean publishing test results, which is not what this is.

1The page has to be one the vendor controls

Pricing pages, plan tables, product documentation and published guides all qualify. A support agent's reply, a post on a community forum, and a screenshot of the product do not, even when they are more specific than anything on the official page.

The reason is reproducibility rather than reliability. A reader can open a pricing page and see for themselves; they cannot open somebody else's support ticket. A record whose rows cannot be re-checked by whoever is reading it is asking for trust it has no way to earn.

2Looking at output would be a different site

Generating a clip and inspecting what comes out would answer several of the questions here faster than waiting for a vendor to publish. It would also produce claims tied to one account, one plan, one day and one version, which is exactly the kind of finding that is stale before anyone reads it.

So the four fields are filled from published pages or left empty. Empty is a real answer here and it appears often: eleven of the sixteen entries have nothing in the machine-readable field.

3One vendor per page, and no aggregate scores

There is no combined rating here and there will not be one. Four fields with different kinds of evidence behind them do not add up to a number, and any weighting used to produce one would be this site's opinion wearing a fact's clothes.

What the table does instead is keep the fields side by side in a fixed order, so a reader applying their own weighting can see every input to it. Somebody who only cares about the visible mark reads one column; somebody planning around metadata reads another and finds it almost entirely empty.

4Omissions are recorded as omissions

When a phrase appears in the paid column of a plan table and not in the free one, the free row says so in those terms. That is weaker than a sentence and stronger than nothing, and collapsing it into either would misrepresent what was read.

Five of the nine free-tier rows on this site are of that kind. Anyone quoting them should carry the qualifier with them, because a plan grid can be redrawn without an announcement and the evidence changes the same day.

5What the delivered file carries

Next to the marking question in “What earns a row”, SceneMixer's pages describe its output like this:

It becomes a marked line — inner voice or narration — so it is heard while the picture shows something worth watching, rather than becoming a shot of somebody thinking.

SceneMixer, AI Dialogue Generator — “What happens to inner monologue from a novel?” (read 2026-10-13)

You own the output — download finished episodes and publish to TikTok, YouTube Shorts, Instagram Reels, or short-drama platforms.

SceneMixer, The AI Micro Drama Production Pipeline — “Where can I publish the episodes?” (read 2026-10-13)

6Sources

The pages each field is read from are listed on the support table, with a date on every row. The rest of the method is on the protocol page, and the companion note is how rows change.